The AI safety company is about to become the AI gold rush, and the people who built the guardrails might be too conflicted to say no.
The Summary
- Anthropic is moving toward a potentially $2 trillion IPO, with prediction markets pricing a valuation above $200B by December at 78.5% YES, while some urge delay over AI safety and regulatory concerns
- Morgan Stanley, likely underwriting the deal, says AI spending won't slow even as it banks almost every company doing the spending, a $2 trillion conflict of interest dressed as analysis
- METR, Anthropic's AI safety evaluator, faces conflict allegations over shared funders, free compute tokens, and revolving door staffing, despite denying direct payments
- Anthropic's head of policy is framing AI safety as a voter issue ahead of midterms, positioning the company to shape regulations just as it seeks public market liquidity
The Signal
Anthropic built its brand on being the "safe" AI lab. Constitutional AI. Responsible scaling. The company that said no when others said ship it. Now that brand is about to meet Wall Street, and the cracks are showing.
The IPO timeline is aggressive. Prediction markets have moved to 78.5% confidence on a $200B+ valuation by year-end. That's not a careful, measured path to public markets. That's a sprint. And the timing matters because Morgan Stanley, positioned to underwrite the offering, is publicly dismissing concerns about an AI spending slowdown while simultaneously banking the hyperscalers, chipmakers, and AI labs driving that spending. It's not analysis. It's salesmanship with a Bloomberg terminal.
"Morgan Stanley banks almost every company doing the spending, then tells you the spending won't stop."
The conflict gets deeper. METR, the nonprofit that evaluates Anthropic's models for dangerous capabilities, is facing viral allegations about its independence. The details matter:
- METR shares major funders with Anthropic, including Open Philanthropy
- Anthropic provides METR with free API credits and compute tokens
- Staff flow between the organizations through a well-worn revolving door
- METR denies taking direct payments, but the indirect ties are structural, not incidental
METR's defense is technically true and functionally weak. They don't take Anthropic's money directly. They just take money from the same pools, use Anthropic's infrastructure for free, and hire from the same talent base. In safety engineering, we'd call that a single point of failure. In finance, we'd call it what it is: conflict of interest.
Meanwhile, Anthropic's Sarah Heck is positioning AI safety as a midterm election issue, pushing the narrative that voters care about AI governance. Translation: Anthropic wants regulatory clarity before the IPO roadshow, and they want to be the ones who helped write it. Smart strategy. Also transparent as glass.
The thesis collision here is brutal. Either:
- AI safety concerns are real and serious, in which case the IPO should be delayed until governance structures actually work
- Or they're manageable PR problems, in which case the "safety-first" brand is marketing
You can't have both. Anthropic is trying anyway.
The Implication
Watch the IPO timing against the election calendar. If Anthropic pushes through before November, it means they believe current regulatory ambiguity is better than post-election clarity. That's a signal about what they expect Congress to do with AI governance.
For builders in the agent economy, this sets the template. Safety theater gets you to IPO. Actual safety slows you down. The market will reward Anthropic's approach or punish it, and every AI lab is watching. If METR's independence questions don't tank the valuation, expect more "independent" evaluators funded by the same philanthropies, using the same compute, hiring the same people. The appearance of oversight without the friction of actual accountability.